Optimized Artificial Neural Network Based On Improved Sparrow Search Algorithm For Short-term Power Load Forecasting

The rapid development of China’s technology and economy has led to a continuous growth in industrial and residential electricity demand,resulting in an increasing number of load types connected to the grid,leading to more serious randomness in the changes in electricity loads.In order to ensure that the power sector can make reasonable power deployment plans,the accuracy and timeliness of power load forecasting needs to be further improved.Therefore,this paper uses an improved sparrow search algorithm to optimise the artificial neural network model to forecast the electricity load with higher accuracy.Firstly,to address the situation that there are some outliers and missing values in the electricity load dataset,this paper introduces the i Forest algorithm to identify and label the outliers,and uses the mean of segmented cubic Elmit interpolation and cubic spline interpolation to replace the outliers and fill in the missing values,thus achieving the purpose of enhancing the electricity load dataset.Secondly,due to the high dimensionality of the original input feature set and the possible existence of factors unrelated to the electricity load,this paper uses a combination of multiple linear regression and factor analysis models to filter and reduce the dimensionality of the data.The multiple linear regression equations are solved and the dependent variables with significant regression coefficients are retained to form the initial input feature set,and the initial feature set is reduced in dimensionality by factor rotation to obtain five common factors,which contain more than 95% of the information in the initial feature set.Finally,ISSA is used to optimize the initial weights and thresholds of the BP neural network and the number of nodes in the hidden layer,training times and learning rates of the LSTM neural network,respectively,to address the shortcomings of each of the two typical artificial neural networks,namely,back propagation neural network and long short-term memory.Meanwhile,ISSA is experimentally compared with the sparrow search algorithm and particle swarm algorithm,and it is proved that ISSA has better performance.Therefore,two neural network models,ISSA-BP and ISSA-LSTM,were constructed to forecast the electricity load,and the prediction results of the two models were compared before and after their respective optimisation,which proved that the prediction results of the optimised models were more accurate.